SceniX develops general-purpose robotics capabilities through advanced hybrid simulation technologies. The company creates digital-twin frameworks, like BoxTwin, that model complex elastoplastic object dynamics directly from video data for adaptive manipulation. Their PhysTwin-Eval system enables high-fidelity, real-to-sim policy evaluation for benchmarking and training robotic learning systems.
Funding
Funding not disclosed
Founders
Product
Problem
Developing robotic learning systems requires extensive training data that accurately reflects real-world object interactions and environmental dynamics, which is often costly and time-consuming to acquire. Existing simulation environments may lack the fidelity needed to effectively transfer learned behaviors to physical robots.
Solution
SceniX is creating a game engine that functions as a high-fidelity world model, enabling rapid development and deployment of robotic learning systems. The engine captures detailed object geometry, appearance, and dynamics, facilitating efficient robotic training and evaluation within a realistic simulated environment. By advancing 3D generative models, SceniX aims to accurately represent real-world object interactions, reducing the gap between simulation and real-world performance for robots.
Target Audience
The primary audience includes robotics researchers, developers, and engineers who require a robust simulation environment for training and testing robotic learning systems.
Features
- High-fidelity simulation of object geometry, appearance, and dynamics
- Advanced 3D generative models for realistic object interactions
- Engine designed to facilitate robotic training and evaluation